@CompleteSkeptic Diego’s thesis is really cool and I hope it wins. Was only like 3 years ago we were comparing Llama outputs from instruction-following (chat) vs multiclass NER, two very different task architectures
@binsquares This is awesome - any chance stealth bypass is in scope for this work? I previously had microVMs setup on a browser agent project but eventually was still blocked in production due to captchas, turnstiles, etc and went with a 3P vendor
@o_kwasniewski@TesterArmy Interested in the migration experience. We’ve been using AI SDK for our browser automation agent and considering the switch to eve
@zeeg@mikehostetler Sonnet 4.6 does as well as Opus 4.7 at our web automation tasks, often more reliable as well. Opus is far more autonomous though (it really doesn’t want to stop)